chore: release v0.6.4

[xdotli]

Release v0.6.4

Cuts 0.6.4 from main. See CHANGELOG.md.

Highlights

• Added — environment + config as run-time axes on bench eval run (feat(eval): bind the environment (S) and config (C) axes at the CLI #790): --state binds the environment (S-axis) per run (inline JSON, registry name@version via the environment registry, or a manifest path — takes precedence over --environment-manifest); --config-override overlays task config (C-axis; inline JSON/YAML/TOML or @file, deep-merged); --config gains a --run-config alias.
• Changed — bench continue is now bench eval continue (feat(cli): move continue under bench eval continue (keep deprecated top-level alias) #800): moved under the eval group (discoverable in bench eval --help); top-level bench continue / continue-batch kept as hidden, deprecated aliases.

Checklist

• Version bumped (0.6.3 → 0.6.4) in pyproject.toml
• CHANGELOG updated ([Unreleased] → 0.6.4)
• uv.lock refreshed
• ruff format + ruff check clean (local)
• CI gate passes (GitHub Actions on this PR)
• Merge, then tag v0.6.4 on main

[greptile-apps]

Greptile Summary

This PR cuts the v0.6.4 release by bumping the version in pyproject.toml and uv.lock (0.6.3 → 0.6.4) and promoting the accumulated [Unreleased] changelog entries into a dated 0.6.4 section.

• Version bump: pyproject.toml and uv.lock updated from 0.6.3 to 0.6.4; all transitive dependency pins in uv.lock are unchanged.
• CHANGELOG: Two feature entries moved from [Unreleased] into 0.6.4 — 2026-06-17 — environment/config run-time axes on bench eval run (feat(eval): bind the environment (S) and config (C) axes at the CLI #790) and the bench continue → bench eval continue rename (feat(cli): move continue under bench eval continue (keep deprecated top-level alias) #800), both with their PR references added.
• The [Unreleased] section is left empty and ready for the next cycle.

Confidence Score: 5/5

Safe to merge — purely a version bump and changelog promotion with no logic changes.

All three changed files are metadata: the version field in pyproject.toml and its mirror in uv.lock move from 0.6.3 to 0.6.4, and CHANGELOG.md promotes two previously-merged feature entries into a dated release section. No source code, configuration logic, or dependency pins are touched.

No files require special attention.

Important Files Changed

Filename	Overview
CHANGELOG.md	Moves two features from [Unreleased] into the 0.6.4 section dated 2026-06-17; adds PR issue references (#790, #800) to both entries.
pyproject.toml	Single-field version bump from 0.6.3 to 0.6.4; no other changes.
uv.lock	Lock file version entry for the benchflow package updated to match pyproject.toml (0.6.3 to 0.6.4); all other dependency pins unchanged.

_{Reviews (2): Last reviewed commit: "docs(changelog): note content-addressed ..." | Re-trigger Greptile}

[bingran-you]

Automation triage update (2026-06-18): release PR is clean and all current checks are green (test, pip-audit, eval-and-judge). I added P1 + review:pending because this still needs a non-author human review before merge under the repo rules. After review, merge/tag v0.6.4 from main.

[bingran-you]

Automation triage (2026-06-19): holding the v0.6.4 release PR even though its current checks are green.

Reason: #804 fixes a release-blocking silent wrong-results path in eval sharding / run-config override handling. Cutting v0.6.4 before #804 is merged would knowingly ship the #790 regression. I added status:blocked; keep this release PR in review until #804 is merged or there is an explicit human decision to defer it, then refresh the release branch/changelog and rerun checks.

[bingran-you]

Automation triage update (2026-06-21): #804 is now merged into main, so the original release blocker is gone, but this release PR still should not merge as-is.

The release branch was cut before #804 and the later main changes, so the next safe step is to refresh/rebase the release PR from current main, update the changelog/version scope to include the newly merged release-bound fixes, rerun CI, then get non-author human review before tagging v0.6.4.

[bingran-you]

Automation triage (2026-06-22): still status:blocked — release should be rebased + changelog-refreshed before merge.

Since this branch was cut, main has advanced substantially. Newly landed on main that belong in the v0.6.4 cut and are not yet in this release branch's CHANGELOG:

• fix(eval): bind resolved S-axis env + C-axis overlay on the sharded and run-config paths #804 (eval sharding / run-config + C-axis overlay correctness — the original release blocker, now merged)
• Make LiteLLM proxy mandatory for routable agents (never bypass; always capture usage/cost/trajectory) #820 (LiteLLM proxy mandatory for routable agents; DOCS_URL boot-crash fix)
• fix(eval): expose context-root on eval run #816 (bench eval run --context-root)
• fix(eval): correct verifier-error resume log #819 (verifier-error resume log fix)
• fix(integration): clear residual greptile findings on the L3 gate #817 (L3 gate greptile residual clear), Add adapter skill #793 (adapter skill)

Decision needed (owner): this is a human release gate — (1) rebase the release branch onto current main, (2) curate the 0.6.4 CHANGELOG to fold in the above, (3) non-author review → squash-merge, (4) tag v0.6.4 + PyPI publish. Not taking automated merge/tag/publish action on a release PR.